src/bin/proxmox-backup-proxy.rs

   1 use std::sync::{Mutex, Arc};
   2 use std::path::{Path, PathBuf};
   3 use std::os::unix::io::AsRawFd;
   4 use std::future::Future;
   5 use std::pin::Pin;
   6
   7 use anyhow::{bail, format_err, Error};
   8 use futures::*;
   9 use http::request::Parts;
  10 use http::Response;
  11 use hyper::{Body, StatusCode};
  12 use hyper::header;
  13 use url::form_urlencoded;
  14
  15 use openssl::ssl::{SslMethod, SslAcceptor, SslFiletype};
  16 use tokio_stream::wrappers::ReceiverStream;
  17 use serde_json::{json, Value};
  18 use http::{Method, HeaderMap};
  19
  20 use proxmox::sys::linux::socket::set_tcp_keepalive;
  21 use proxmox::tools::fs::CreateOptions;
  22 use proxmox_lang::try_block;
  23 use proxmox_router::{RpcEnvironment, RpcEnvironmentType, UserInformation};
  24 use proxmox_http::client::{RateLimitedStream, ShareableRateLimit};
  25 use proxmox_sys::{task_log, task_warn};
  26 use proxmox_sys::logrotate::LogRotate;
  27
  28 use pbs_datastore::DataStore;
  29
  30 use proxmox_rest_server::{
  31     rotate_task_log_archive, extract_cookie , AuthError, ApiConfig, RestServer, RestEnvironment,
  32     ServerAdapter, WorkerTask, cleanup_old_tasks,
  33 };
  34
  35 use proxmox_backup::rrd_cache::{
  36     initialize_rrd_cache, rrd_update_gauge, rrd_update_derive, rrd_sync_journal,
  37 };
  38 use proxmox_backup::{
  39     TRAFFIC_CONTROL_CACHE,
  40     server::{
  41         auth::check_pbs_auth,
  42         jobstate::{
  43             self,
  44             Job,
  45         },
  46     },
  47 };
  48
  49 use pbs_buildcfg::configdir;
  50 use proxmox_time::{compute_next_event, parse_calendar_event};
  51
  52 use pbs_api_types::{
  53     Authid, TapeBackupJobConfig, VerificationJobConfig, SyncJobConfig, DataStoreConfig,
  54     PruneOptions,
  55 };
  56
  57 use proxmox_rest_server::daemon;
  58
  59 use proxmox_backup::server;
  60 use proxmox_backup::auth_helpers::*;
  61 use proxmox_backup::tools::{
  62     PROXMOX_BACKUP_TCP_KEEPALIVE_TIME,
  63     disks::{
  64         DiskManage,
  65         zfs_pool_stats,
  66         get_pool_from_dataset,
  67     },
  68 };
  69
  70
  71 use proxmox_backup::api2::pull::do_sync_job;
  72 use proxmox_backup::api2::tape::backup::do_tape_backup_job;
  73 use proxmox_backup::server::do_verification_job;
  74 use proxmox_backup::server::do_prune_job;
  75
  76 fn main() -> Result<(), Error> {
  77     proxmox_backup::tools::setup_safe_path_env();
  78
  79     let backup_uid = pbs_config::backup_user()?.uid;
  80     let backup_gid = pbs_config::backup_group()?.gid;
  81     let running_uid = nix::unistd::Uid::effective();
  82     let running_gid = nix::unistd::Gid::effective();
  83
  84     if running_uid != backup_uid || running_gid != backup_gid {
  85         bail!("proxy not running as backup user or group (got uid {} gid {})", running_uid, running_gid);
  86     }
  87
  88     pbs_runtime::main(run())
  89 }
  90
  91
  92 struct ProxmoxBackupProxyAdapter;
  93
  94 impl ServerAdapter for ProxmoxBackupProxyAdapter {
  95
  96     fn get_index(
  97         &self,
  98         env: RestEnvironment,
  99         parts: Parts,
 100     ) -> Pin<Box<dyn Future<Output = Response<Body>> + Send>> {
 101         Box::pin(get_index_future(env, parts))
 102     }
 103
 104     fn check_auth<'a>(
 105         &'a self,
 106         headers: &'a HeaderMap,
 107         method: &'a Method,
 108     ) -> Pin<Box<dyn Future<Output = Result<(String, Box<dyn UserInformation + Sync + Send>), AuthError>> + Send + 'a>> {
 109         Box::pin(async move {
 110             check_pbs_auth(headers, method).await
 111         })
 112     }
 113 }
 114
 115 fn extract_lang_header(headers: &http::HeaderMap) -> Option<String> {
 116     if let Some(Ok(cookie)) = headers.get("COOKIE").map(|v| v.to_str()) {
 117         return extract_cookie(cookie, "PBSLangCookie");
 118     }
 119     None
 120 }
 121
 122 async fn get_index_future(
 123     env: RestEnvironment,
 124     parts: Parts,
 125 ) -> Response<Body> {
 126
 127     let auth_id = env.get_auth_id();
 128     let api = env.api_config();
 129     let language = extract_lang_header(&parts.headers);
 130
 131      // fixme: make all IO async
 132
 133     let (userid, csrf_token) = match auth_id {
 134         Some(auth_id) => {
 135             let auth_id = auth_id.parse::<Authid>();
 136             match auth_id {
 137                 Ok(auth_id) if !auth_id.is_token() => {
 138                     let userid = auth_id.user().clone();
 139                     let new_csrf_token = assemble_csrf_prevention_token(csrf_secret(), &userid);
 140                     (Some(userid), Some(new_csrf_token))
 141                 }
 142                 _ => (None, None)
 143             }
 144         }
 145         None => (None, None),
 146     };
 147
 148     let nodename = proxmox::tools::nodename();
 149     let user = userid.as_ref().map(|u| u.as_str()).unwrap_or("");
 150
 151     let csrf_token = csrf_token.unwrap_or_else(|| String::from(""));
 152
 153     let mut debug = false;
 154     let mut template_file = "index";
 155
 156     if let Some(query_str) = parts.uri.query() {
 157         for (k, v) in form_urlencoded::parse(query_str.as_bytes()).into_owned() {
 158             if k == "debug" && v != "0" && v != "false" {
 159                 debug = true;
 160             } else if k == "console" {
 161                 template_file = "console";
 162             }
 163         }
 164     }
 165
 166     let mut lang = String::from("");
 167     if let Some(language) = language {
 168         if Path::new(&format!("/usr/share/pbs-i18n/pbs-lang-{}.js", language)).exists() {
 169             lang = language;
 170         }
 171     }
 172
 173     let data = json!({
 174         "NodeName": nodename,
 175         "UserName": user,
 176         "CSRFPreventionToken": csrf_token,
 177         "language": lang,
 178         "debug": debug,
 179     });
 180
 181     let (ct, index) = match api.render_template(template_file, &data) {
 182         Ok(index) => ("text/html", index),
 183         Err(err) => ("text/plain", format!("Error rendering template: {}", err)),
 184     };
 185
 186     let mut resp = Response::builder()
 187         .status(StatusCode::OK)
 188         .header(header::CONTENT_TYPE, ct)
 189         .body(index.into())
 190         .unwrap();
 191
 192     if let Some(userid) = userid {
 193         resp.extensions_mut().insert(Authid::from((userid, None)));
 194     }
 195
 196     resp
 197 }
 198
 199 async fn run() -> Result<(), Error> {
 200     if let Err(err) = syslog::init(
 201         syslog::Facility::LOG_DAEMON,
 202         log::LevelFilter::Info,
 203         Some("proxmox-backup-proxy")) {
 204         bail!("unable to inititialize syslog - {}", err);
 205     }
 206
 207     // Note: To debug early connection error use
 208     // PROXMOX_DEBUG=1 ./target/release/proxmox-backup-proxy
 209     let debug = std::env::var("PROXMOX_DEBUG").is_ok();
 210
 211     let _ = public_auth_key(); // load with lazy_static
 212     let _ = csrf_secret(); // load with lazy_static
 213
 214     let rrd_cache = initialize_rrd_cache()?;
 215     rrd_cache.apply_journal()?;
 216
 217     let mut config = ApiConfig::new(
 218         pbs_buildcfg::JS_DIR,
 219         &proxmox_backup::api2::ROUTER,
 220         RpcEnvironmentType::PUBLIC,
 221         ProxmoxBackupProxyAdapter,
 222     )?;
 223
 224     config.add_alias("novnc", "/usr/share/novnc-pve");
 225     config.add_alias("extjs", "/usr/share/javascript/extjs");
 226     config.add_alias("qrcodejs", "/usr/share/javascript/qrcodejs");
 227     config.add_alias("fontawesome", "/usr/share/fonts-font-awesome");
 228     config.add_alias("xtermjs", "/usr/share/pve-xtermjs");
 229     config.add_alias("locale", "/usr/share/pbs-i18n");
 230     config.add_alias("widgettoolkit", "/usr/share/javascript/proxmox-widget-toolkit");
 231     config.add_alias("docs", "/usr/share/doc/proxmox-backup/html");
 232
 233     let mut indexpath = PathBuf::from(pbs_buildcfg::JS_DIR);
 234     indexpath.push("index.hbs");
 235     config.register_template("index", &indexpath)?;
 236     config.register_template("console", "/usr/share/pve-xtermjs/index.html.hbs")?;
 237
 238     let backup_user = pbs_config::backup_user()?;
 239     let mut commando_sock = proxmox_rest_server::CommandSocket::new(proxmox_rest_server::our_ctrl_sock(), backup_user.gid);
 240
 241     let dir_opts = CreateOptions::new().owner(backup_user.uid).group(backup_user.gid);
 242     let file_opts = CreateOptions::new().owner(backup_user.uid).group(backup_user.gid);
 243
 244     config.enable_access_log(
 245         pbs_buildcfg::API_ACCESS_LOG_FN,
 246         Some(dir_opts.clone()),
 247         Some(file_opts.clone()),
 248         &mut commando_sock,
 249     )?;
 250
 251     config.enable_auth_log(
 252         pbs_buildcfg::API_AUTH_LOG_FN,
 253         Some(dir_opts.clone()),
 254         Some(file_opts.clone()),
 255         &mut commando_sock,
 256     )?;
 257
 258     let rest_server = RestServer::new(config);
 259     proxmox_rest_server::init_worker_tasks(pbs_buildcfg::PROXMOX_BACKUP_LOG_DIR_M!().into(), file_opts.clone())?;
 260
 261     //openssl req -x509 -newkey rsa:4096 -keyout /etc/proxmox-backup/proxy.key -out /etc/proxmox-backup/proxy.pem -nodes
 262
 263     // we build the initial acceptor here as we cannot start if this fails
 264     let acceptor = make_tls_acceptor()?;
 265     let acceptor = Arc::new(Mutex::new(acceptor));
 266
 267     // to renew the acceptor we just add a command-socket handler
 268     commando_sock.register_command(
 269         "reload-certificate".to_string(),
 270         {
 271             let acceptor = Arc::clone(&acceptor);
 272             move |_value| -> Result<_, Error> {
 273                 log::info!("reloading certificate");
 274                 match make_tls_acceptor() {
 275                     Err(err) => log::error!("error reloading certificate: {}", err),
 276                     Ok(new_acceptor) => {
 277                         let mut guard = acceptor.lock().unwrap();
 278                         *guard = new_acceptor;
 279                     }
 280                 }
 281                 Ok(Value::Null)
 282             }
 283         },
 284     )?;
 285
 286     // to remove references for not configured datastores
 287     commando_sock.register_command(
 288         "datastore-removed".to_string(),
 289         |_value| {
 290             if let Err(err) = DataStore::remove_unused_datastores() {
 291                 log::error!("could not refresh datastores: {}", err);
 292             }
 293             Ok(Value::Null)
 294         }
 295     )?;
 296
 297     let server = daemon::create_daemon(
 298         ([0,0,0,0,0,0,0,0], 8007).into(),
 299         move |listener| {
 300
 301             let connections = accept_connections(listener, acceptor, debug);
 302             let connections = hyper::server::accept::from_stream(ReceiverStream::new(connections));
 303
 304             Ok(async {
 305                 daemon::systemd_notify(daemon::SystemdNotify::Ready)?;
 306
 307                 hyper::Server::builder(connections)
 308                     .serve(rest_server)
 309                     .with_graceful_shutdown(proxmox_rest_server::shutdown_future())
 310                     .map_err(Error::from)
 311                     .await
 312             })
 313         },
 314     );
 315
 316     proxmox_rest_server::write_pid(pbs_buildcfg::PROXMOX_BACKUP_PROXY_PID_FN)?;
 317
 318     let init_result: Result<(), Error> = try_block!({
 319         proxmox_rest_server::register_task_control_commands(&mut commando_sock)?;
 320         commando_sock.spawn()?;
 321         proxmox_rest_server::catch_shutdown_signal()?;
 322         proxmox_rest_server::catch_reload_signal()?;
 323         Ok(())
 324     });
 325
 326     if let Err(err) = init_result {
 327         bail!("unable to start daemon - {}", err);
 328     }
 329
 330     start_task_scheduler();
 331     start_stat_generator();
 332     start_traffic_control_updater();
 333
 334     server.await?;
 335     log::info!("server shutting down, waiting for active workers to complete");
 336     proxmox_rest_server::last_worker_future().await?;
 337     log::info!("done - exit server");
 338
 339     Ok(())
 340 }
 341
 342 fn make_tls_acceptor() -> Result<SslAcceptor, Error> {
 343     let key_path = configdir!("/proxy.key");
 344     let cert_path = configdir!("/proxy.pem");
 345
 346     let mut acceptor = SslAcceptor::mozilla_intermediate_v5(SslMethod::tls()).unwrap();
 347     acceptor.set_private_key_file(key_path, SslFiletype::PEM)
 348         .map_err(|err| format_err!("unable to read proxy key {} - {}", key_path, err))?;
 349     acceptor.set_certificate_chain_file(cert_path)
 350         .map_err(|err| format_err!("unable to read proxy cert {} - {}", cert_path, err))?;
 351     acceptor.check_private_key().unwrap();
 352
 353     Ok(acceptor.build())
 354 }
 355
 356 type ClientStreamResult =
 357     Result<std::pin::Pin<Box<tokio_openssl::SslStream<RateLimitedStream<tokio::net::TcpStream>>>>, Error>;
 358 const MAX_PENDING_ACCEPTS: usize = 1024;
 359
 360 fn accept_connections(
 361     listener: tokio::net::TcpListener,
 362     acceptor: Arc<Mutex<openssl::ssl::SslAcceptor>>,
 363     debug: bool,
 364 ) -> tokio::sync::mpsc::Receiver<ClientStreamResult> {
 365
 366     let (sender, receiver) = tokio::sync::mpsc::channel(MAX_PENDING_ACCEPTS);
 367
 368     tokio::spawn(accept_connection(listener, acceptor, debug, sender));
 369
 370     receiver
 371 }
 372
 373 async fn accept_connection(
 374     listener: tokio::net::TcpListener,
 375     acceptor: Arc<Mutex<openssl::ssl::SslAcceptor>>,
 376     debug: bool,
 377     sender: tokio::sync::mpsc::Sender<ClientStreamResult>,
 378 ) {
 379     let accept_counter = Arc::new(());
 380
 381     loop {
 382         let (sock, _addr) = match listener.accept().await {
 383             Ok(conn) => conn,
 384             Err(err) =>  {
 385                 eprintln!("error accepting tcp connection: {}", err);
 386                 continue;
 387             }
 388         };
 389
 390         sock.set_nodelay(true).unwrap();
 391         let _ = set_tcp_keepalive(sock.as_raw_fd(), PROXMOX_BACKUP_TCP_KEEPALIVE_TIME);
 392
 393         let peer = sock.peer_addr().ok();
 394         let sock = RateLimitedStream::with_limiter_update_cb(sock, move || lookup_rate_limiter(peer));
 395
 396         let ssl = { // limit acceptor_guard scope
 397             // Acceptor can be reloaded using the command socket "reload-certificate" command
 398             let acceptor_guard = acceptor.lock().unwrap();
 399
 400             match openssl::ssl::Ssl::new(acceptor_guard.context()) {
 401                 Ok(ssl) => ssl,
 402                 Err(err) => {
 403                     eprintln!("failed to create Ssl object from Acceptor context - {}", err);
 404                     continue;
 405                 },
 406             }
 407         };
 408
 409         let stream = match tokio_openssl::SslStream::new(ssl, sock) {
 410             Ok(stream) => stream,
 411             Err(err) => {
 412                 eprintln!("failed to create SslStream using ssl and connection socket - {}", err);
 413                 continue;
 414             },
 415         };
 416
 417         let mut stream = Box::pin(stream);
 418         let sender = sender.clone();
 419
 420         if Arc::strong_count(&accept_counter) > MAX_PENDING_ACCEPTS {
 421             eprintln!("connection rejected - to many open connections");
 422             continue;
 423         }
 424
 425         let accept_counter = Arc::clone(&accept_counter);
 426         tokio::spawn(async move {
 427             let accept_future = tokio::time::timeout(
 428                 Duration::new(10, 0), stream.as_mut().accept());
 429
 430             let result = accept_future.await;
 431
 432             match result {
 433                 Ok(Ok(())) => {
 434                     if sender.send(Ok(stream)).await.is_err() && debug {
 435                         eprintln!("detect closed connection channel");
 436                     }
 437                 }
 438                 Ok(Err(err)) => {
 439                     if debug {
 440                         eprintln!("https handshake failed - {}", err);
 441                     }
 442                 }
 443                 Err(_) => {
 444                     if debug {
 445                         eprintln!("https handshake timeout");
 446                     }
 447                 }
 448             }
 449
 450             drop(accept_counter); // decrease reference count
 451         });
 452     }
 453 }
 454
 455 fn start_stat_generator() {
 456     let abort_future = proxmox_rest_server::shutdown_future();
 457     let future = Box::pin(run_stat_generator());
 458     let task = futures::future::select(future, abort_future);
 459     tokio::spawn(task.map(|_| ()));
 460 }
 461
 462 fn start_task_scheduler() {
 463     let abort_future = proxmox_rest_server::shutdown_future();
 464     let future = Box::pin(run_task_scheduler());
 465     let task = futures::future::select(future, abort_future);
 466     tokio::spawn(task.map(|_| ()));
 467 }
 468
 469 fn start_traffic_control_updater() {
 470     let abort_future = proxmox_rest_server::shutdown_future();
 471     let future = Box::pin(run_traffic_control_updater());
 472     let task = futures::future::select(future, abort_future);
 473     tokio::spawn(task.map(|_| ()));
 474 }
 475
 476 use std::time::{SystemTime, Instant, Duration, UNIX_EPOCH};
 477
 478 fn next_minute() -> Result<Instant, Error> {
 479     let now = SystemTime::now();
 480     let epoch_now = now.duration_since(UNIX_EPOCH)?;
 481     let epoch_next = Duration::from_secs((epoch_now.as_secs()/60  + 1)*60);
 482     Ok(Instant::now() + epoch_next - epoch_now)
 483 }
 484
 485 async fn run_task_scheduler() {
 486
 487     let mut count: usize = 0;
 488
 489     loop {
 490         count += 1;
 491
 492         let delay_target = match next_minute() {  // try to run very minute
 493             Ok(d) => d,
 494             Err(err) => {
 495                 eprintln!("task scheduler: compute next minute failed - {}", err);
 496                 tokio::time::sleep_until(tokio::time::Instant::from_std(Instant::now() + Duration::from_secs(60))).await;
 497                 continue;
 498             }
 499         };
 500
 501         if count > 2 { // wait 1..2 minutes before starting
 502             match schedule_tasks().catch_unwind().await {
 503                 Err(panic) => {
 504                     match panic.downcast::<&str>() {
 505                         Ok(msg) => {
 506                             eprintln!("task scheduler panic: {}", msg);
 507                         }
 508                         Err(_) => {
 509                             eprintln!("task scheduler panic - unknown type");
 510                         }
 511                     }
 512                 }
 513                 Ok(Err(err)) => {
 514                     eprintln!("task scheduler failed - {:?}", err);
 515                 }
 516                 Ok(Ok(_)) => {}
 517             }
 518         }
 519
 520         tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
 521     }
 522 }
 523
 524 async fn schedule_tasks() -> Result<(), Error> {
 525
 526     schedule_datastore_garbage_collection().await;
 527     schedule_datastore_prune().await;
 528     schedule_datastore_sync_jobs().await;
 529     schedule_datastore_verify_jobs().await;
 530     schedule_tape_backup_jobs().await;
 531     schedule_task_log_rotate().await;
 532
 533     Ok(())
 534 }
 535
 536 async fn schedule_datastore_garbage_collection() {
 537
 538     let config = match pbs_config::datastore::config() {
 539         Err(err) => {
 540             eprintln!("unable to read datastore config - {}", err);
 541             return;
 542         }
 543         Ok((config, _digest)) => config,
 544     };
 545
 546     for (store, (_, store_config)) in config.sections {
 547         let datastore = match DataStore::lookup_datastore(&store) {
 548             Ok(datastore) => datastore,
 549             Err(err) => {
 550                 eprintln!("lookup_datastore failed - {}", err);
 551                 continue;
 552             }
 553         };
 554
 555         let store_config: DataStoreConfig = match serde_json::from_value(store_config) {
 556             Ok(c) => c,
 557             Err(err) => {
 558                 eprintln!("datastore config from_value failed - {}", err);
 559                 continue;
 560             }
 561         };
 562
 563         let event_str = match store_config.gc_schedule {
 564             Some(event_str) => event_str,
 565             None => continue,
 566         };
 567
 568         let event = match parse_calendar_event(&event_str) {
 569             Ok(event) => event,
 570             Err(err) => {
 571                 eprintln!("unable to parse schedule '{}' - {}", event_str, err);
 572                 continue;
 573             }
 574         };
 575
 576         if datastore.garbage_collection_running() { continue; }
 577
 578         let worker_type = "garbage_collection";
 579
 580         let last = match jobstate::last_run_time(worker_type, &store) {
 581             Ok(time) => time,
 582             Err(err) => {
 583                 eprintln!("could not get last run time of {} {}: {}", worker_type, store, err);
 584                 continue;
 585             }
 586         };
 587
 588         let next = match compute_next_event(&event, last, false) {
 589             Ok(Some(next)) => next,
 590             Ok(None) => continue,
 591             Err(err) => {
 592                 eprintln!("compute_next_event for '{}' failed - {}", event_str, err);
 593                 continue;
 594             }
 595         };
 596
 597         let now = proxmox_time::epoch_i64();
 598
 599         if next > now  { continue; }
 600
 601         let job = match Job::new(worker_type, &store) {
 602             Ok(job) => job,
 603             Err(_) => continue, // could not get lock
 604         };
 605
 606         let auth_id = Authid::root_auth_id();
 607
 608         if let Err(err) = crate::server::do_garbage_collection_job(job, datastore, auth_id, Some(event_str), false) {
 609             eprintln!("unable to start garbage collection job on datastore {} - {}", store, err);
 610         }
 611     }
 612 }
 613
 614 async fn schedule_datastore_prune() {
 615
 616     let config = match pbs_config::datastore::config() {
 617         Err(err) => {
 618             eprintln!("unable to read datastore config - {}", err);
 619             return;
 620         }
 621         Ok((config, _digest)) => config,
 622     };
 623
 624     for (store, (_, store_config)) in config.sections {
 625
 626         let store_config: DataStoreConfig = match serde_json::from_value(store_config) {
 627             Ok(c) => c,
 628             Err(err) => {
 629                 eprintln!("datastore '{}' config from_value failed - {}", store, err);
 630                 continue;
 631             }
 632         };
 633
 634         let event_str = match store_config.prune_schedule {
 635             Some(event_str) => event_str,
 636             None => continue,
 637         };
 638
 639         let prune_options = PruneOptions {
 640             keep_last: store_config.keep_last,
 641             keep_hourly: store_config.keep_hourly,
 642             keep_daily:  store_config.keep_daily,
 643             keep_weekly: store_config.keep_weekly,
 644             keep_monthly: store_config.keep_monthly,
 645             keep_yearly: store_config.keep_yearly,
 646         };
 647
 648         if !pbs_datastore::prune::keeps_something(&prune_options) { // no prune settings - keep all
 649             continue;
 650         }
 651
 652         let worker_type = "prune";
 653         if check_schedule(worker_type, &event_str, &store) {
 654             let job = match Job::new(worker_type, &store) {
 655                 Ok(job) => job,
 656                 Err(_) => continue, // could not get lock
 657             };
 658
 659             let auth_id = Authid::root_auth_id().clone();
 660             if let Err(err) = do_prune_job(job, prune_options, store.clone(), &auth_id, Some(event_str)) {
 661                 eprintln!("unable to start datastore prune job {} - {}", &store, err);
 662             }
 663         };
 664     }
 665 }
 666
 667 async fn schedule_datastore_sync_jobs() {
 668
 669
 670     let config = match pbs_config::sync::config() {
 671         Err(err) => {
 672             eprintln!("unable to read sync job config - {}", err);
 673             return;
 674         }
 675         Ok((config, _digest)) => config,
 676     };
 677
 678     for (job_id, (_, job_config)) in config.sections {
 679         let job_config: SyncJobConfig = match serde_json::from_value(job_config) {
 680             Ok(c) => c,
 681             Err(err) => {
 682                 eprintln!("sync job config from_value failed - {}", err);
 683                 continue;
 684             }
 685         };
 686
 687         let event_str = match job_config.schedule {
 688             Some(ref event_str) => event_str.clone(),
 689             None => continue,
 690         };
 691
 692         let worker_type = "syncjob";
 693         if check_schedule(worker_type, &event_str, &job_id) {
 694             let job = match Job::new(worker_type, &job_id) {
 695                 Ok(job) => job,
 696                 Err(_) => continue, // could not get lock
 697             };
 698
 699             let auth_id = Authid::root_auth_id().clone();
 700             if let Err(err) = do_sync_job(job, job_config, &auth_id, Some(event_str), false) {
 701                 eprintln!("unable to start datastore sync job {} - {}", &job_id, err);
 702             }
 703         };
 704     }
 705 }
 706
 707 async fn schedule_datastore_verify_jobs() {
 708
 709     let config = match pbs_config::verify::config() {
 710         Err(err) => {
 711             eprintln!("unable to read verification job config - {}", err);
 712             return;
 713         }
 714         Ok((config, _digest)) => config,
 715     };
 716     for (job_id, (_, job_config)) in config.sections {
 717         let job_config: VerificationJobConfig = match serde_json::from_value(job_config) {
 718             Ok(c) => c,
 719             Err(err) => {
 720                 eprintln!("verification job config from_value failed - {}", err);
 721                 continue;
 722             }
 723         };
 724         let event_str = match job_config.schedule {
 725             Some(ref event_str) => event_str.clone(),
 726             None => continue,
 727         };
 728
 729         let worker_type = "verificationjob";
 730         let auth_id = Authid::root_auth_id().clone();
 731         if check_schedule(worker_type, &event_str, &job_id) {
 732             let job = match Job::new(&worker_type, &job_id) {
 733                 Ok(job) => job,
 734                 Err(_) => continue, // could not get lock
 735             };
 736             if let Err(err) = do_verification_job(job, job_config, &auth_id, Some(event_str), false) {
 737                 eprintln!("unable to start datastore verification job {} - {}", &job_id, err);
 738             }
 739         };
 740     }
 741 }
 742
 743 async fn schedule_tape_backup_jobs() {
 744
 745     let config = match pbs_config::tape_job::config() {
 746         Err(err) => {
 747             eprintln!("unable to read tape job config - {}", err);
 748             return;
 749         }
 750         Ok((config, _digest)) => config,
 751     };
 752     for (job_id, (_, job_config)) in config.sections {
 753         let job_config: TapeBackupJobConfig = match serde_json::from_value(job_config) {
 754             Ok(c) => c,
 755             Err(err) => {
 756                 eprintln!("tape backup job config from_value failed - {}", err);
 757                 continue;
 758             }
 759         };
 760         let event_str = match job_config.schedule {
 761             Some(ref event_str) => event_str.clone(),
 762             None => continue,
 763         };
 764
 765         let worker_type = "tape-backup-job";
 766         let auth_id = Authid::root_auth_id().clone();
 767         if check_schedule(worker_type, &event_str, &job_id) {
 768             let job = match Job::new(&worker_type, &job_id) {
 769                 Ok(job) => job,
 770                 Err(_) => continue, // could not get lock
 771             };
 772             if let Err(err) = do_tape_backup_job(job, job_config.setup, &auth_id, Some(event_str), false) {
 773                 eprintln!("unable to start tape backup job {} - {}", &job_id, err);
 774             }
 775         };
 776     }
 777 }
 778
 779
 780 async fn schedule_task_log_rotate() {
 781
 782     let worker_type = "logrotate";
 783     let job_id = "access-log_and_task-archive";
 784
 785     // schedule daily at 00:00 like normal logrotate
 786     let schedule = "00:00";
 787
 788     if !check_schedule(worker_type, schedule, job_id) {
 789         // if we never ran the rotation, schedule instantly
 790         match jobstate::JobState::load(worker_type, job_id) {
 791             Ok(state) => match state {
 792                 jobstate::JobState::Created { .. } => {},
 793                 _ => return,
 794             },
 795             _ => return,
 796         }
 797     }
 798
 799     let mut job = match Job::new(worker_type, job_id) {
 800         Ok(job) => job,
 801         Err(_) => return, // could not get lock
 802     };
 803
 804     if let Err(err) = WorkerTask::new_thread(
 805         worker_type,
 806         None,
 807         Authid::root_auth_id().to_string(),
 808         false,
 809         move |worker| {
 810             job.start(&worker.upid().to_string())?;
 811             task_log!(worker, "starting task log rotation");
 812
 813             let result = try_block!({
 814                 let max_size = 512 * 1024 - 1; // an entry has ~ 100b, so > 5000 entries/file
 815                 let max_files = 20; // times twenty files gives > 100000 task entries
 816
 817                 let user = pbs_config::backup_user()?;
 818                 let options = proxmox::tools::fs::CreateOptions::new()
 819                     .owner(user.uid)
 820                     .group(user.gid);
 821
 822                 let has_rotated = rotate_task_log_archive(
 823                     max_size,
 824                     true,
 825                     Some(max_files),
 826                     Some(options.clone()),
 827                 )?;
 828
 829                 if has_rotated {
 830                     task_log!(worker, "task log archive was rotated");
 831                 } else {
 832                     task_log!(worker, "task log archive was not rotated");
 833                 }
 834
 835                 let max_size = 32 * 1024 * 1024 - 1;
 836                 let max_files = 14;
 837
 838
 839                 let mut logrotate = LogRotate::new(
 840                     pbs_buildcfg::API_ACCESS_LOG_FN,
 841                     true,
 842                     Some(max_files),
 843                     Some(options.clone()),
 844                 )?;
 845
 846                 if logrotate.rotate(max_size)? {
 847                     println!("rotated access log, telling daemons to re-open log file");
 848                     pbs_runtime::block_on(command_reopen_access_logfiles())?;
 849                     task_log!(worker, "API access log was rotated");
 850                 } else {
 851                     task_log!(worker, "API access log was not rotated");
 852                 }
 853
 854                 let mut logrotate = LogRotate::new(
 855                     pbs_buildcfg::API_AUTH_LOG_FN,
 856                     true,
 857                     Some(max_files),
 858                     Some(options),
 859                 )?;
 860
 861                 if logrotate.rotate(max_size)? {
 862                     println!("rotated auth log, telling daemons to re-open log file");
 863                     pbs_runtime::block_on(command_reopen_auth_logfiles())?;
 864                     task_log!(worker, "API authentication log was rotated");
 865                 } else {
 866                     task_log!(worker, "API authentication log was not rotated");
 867                 }
 868
 869                 if has_rotated {
 870                     task_log!(worker, "cleaning up old task logs");
 871                     if let Err(err) = cleanup_old_tasks(true) {
 872                         task_warn!(worker, "could not completely cleanup old tasks: {}", err);
 873                     }
 874                 }
 875
 876                 Ok(())
 877             });
 878
 879             let status = worker.create_state(&result);
 880
 881             if let Err(err) = job.finish(status) {
 882                 eprintln!("could not finish job state for {}: {}", worker_type, err);
 883             }
 884
 885             result
 886         },
 887     ) {
 888         eprintln!("unable to start task log rotation: {}", err);
 889     }
 890
 891 }
 892
 893 async fn command_reopen_access_logfiles() -> Result<(), Error> {
 894     // only care about the most recent daemon instance for each, proxy & api, as other older ones
 895     // should not respond to new requests anyway, but only finish their current one and then exit.
 896     let sock = proxmox_rest_server::our_ctrl_sock();
 897     let f1 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-access-log-reopen\"}\n");
 898
 899     let pid = proxmox_rest_server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?;
 900     let sock = proxmox_rest_server::ctrl_sock_from_pid(pid);
 901     let f2 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-access-log-reopen\"}\n");
 902
 903     match futures::join!(f1, f2) {
 904         (Err(e1), Err(e2)) => Err(format_err!("reopen commands failed, proxy: {}; api: {}", e1, e2)),
 905         (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)),
 906         (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)),
 907         _ => Ok(()),
 908     }
 909 }
 910
 911 async fn command_reopen_auth_logfiles() -> Result<(), Error> {
 912     // only care about the most recent daemon instance for each, proxy & api, as other older ones
 913     // should not respond to new requests anyway, but only finish their current one and then exit.
 914     let sock = proxmox_rest_server::our_ctrl_sock();
 915     let f1 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n");
 916
 917     let pid = proxmox_rest_server::read_pid(pbs_buildcfg::PROXMOX_BACKUP_API_PID_FN)?;
 918     let sock = proxmox_rest_server::ctrl_sock_from_pid(pid);
 919     let f2 = proxmox_rest_server::send_raw_command(sock, "{\"command\":\"api-auth-log-reopen\"}\n");
 920
 921     match futures::join!(f1, f2) {
 922         (Err(e1), Err(e2)) => Err(format_err!("reopen commands failed, proxy: {}; api: {}", e1, e2)),
 923         (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)),
 924         (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)),
 925         _ => Ok(()),
 926     }
 927 }
 928
 929 async fn run_stat_generator() {
 930
 931     loop {
 932         let delay_target = Instant::now() +  Duration::from_secs(10);
 933
 934         generate_host_stats().await;
 935
 936         rrd_sync_journal();
 937
 938         tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
 939
 940      }
 941
 942 }
 943
 944 async fn generate_host_stats() {
 945     match tokio::task::spawn_blocking(generate_host_stats_sync).await {
 946         Ok(()) => (),
 947         Err(err) => log::error!("generate_host_stats paniced: {}", err),
 948     }
 949 }
 950
 951 fn generate_host_stats_sync() {
 952     use proxmox::sys::linux::procfs::{
 953         read_meminfo, read_proc_stat, read_proc_net_dev, read_loadavg};
 954
 955     match read_proc_stat() {
 956         Ok(stat) => {
 957             rrd_update_gauge("host/cpu", stat.cpu);
 958             rrd_update_gauge("host/iowait", stat.iowait_percent);
 959         }
 960         Err(err) => {
 961             eprintln!("read_proc_stat failed - {}", err);
 962         }
 963     }
 964
 965     match read_meminfo() {
 966         Ok(meminfo) => {
 967             rrd_update_gauge("host/memtotal", meminfo.memtotal as f64);
 968             rrd_update_gauge("host/memused", meminfo.memused as f64);
 969             rrd_update_gauge("host/swaptotal", meminfo.swaptotal as f64);
 970             rrd_update_gauge("host/swapused", meminfo.swapused as f64);
 971         }
 972         Err(err) => {
 973             eprintln!("read_meminfo failed - {}", err);
 974         }
 975     }
 976
 977     match read_proc_net_dev() {
 978         Ok(netdev) => {
 979             use pbs_config::network::is_physical_nic;
 980             let mut netin = 0;
 981             let mut netout = 0;
 982             for item in netdev {
 983                 if !is_physical_nic(&item.device) { continue; }
 984                 netin += item.receive;
 985                 netout += item.send;
 986             }
 987             rrd_update_derive("host/netin", netin as f64);
 988             rrd_update_derive("host/netout", netout as f64);
 989         }
 990         Err(err) => {
 991             eprintln!("read_prox_net_dev failed - {}", err);
 992         }
 993     }
 994
 995     match read_loadavg() {
 996         Ok(loadavg) => {
 997             rrd_update_gauge("host/loadavg", loadavg.0 as f64);
 998         }
 999         Err(err) => {
1000             eprintln!("read_loadavg failed - {}", err);
1001         }
1002     }
1003
1004     let disk_manager = DiskManage::new();
1005
1006     gather_disk_stats(disk_manager.clone(), Path::new("/"), "host");
1007
1008     match pbs_config::datastore::config() {
1009         Ok((config, _)) => {
1010             let datastore_list: Vec<DataStoreConfig> =
1011                 config.convert_to_typed_array("datastore").unwrap_or_default();
1012
1013             for config in datastore_list {
1014
1015                 let rrd_prefix = format!("datastore/{}", config.name);
1016                 let path = std::path::Path::new(&config.path);
1017                 gather_disk_stats(disk_manager.clone(), path, &rrd_prefix);
1018             }
1019         }
1020         Err(err) => {
1021             eprintln!("read datastore config failed - {}", err);
1022         }
1023     }
1024 }
1025
1026 fn check_schedule(worker_type: &str, event_str: &str, id: &str) -> bool {
1027     let event = match parse_calendar_event(event_str) {
1028         Ok(event) => event,
1029         Err(err) => {
1030             eprintln!("unable to parse schedule '{}' - {}", event_str, err);
1031             return false;
1032         }
1033     };
1034
1035     let last = match jobstate::last_run_time(worker_type, &id) {
1036         Ok(time) => time,
1037         Err(err) => {
1038             eprintln!("could not get last run time of {} {}: {}", worker_type, id, err);
1039             return false;
1040         }
1041     };
1042
1043     let next = match compute_next_event(&event, last, false) {
1044         Ok(Some(next)) => next,
1045         Ok(None) => return false,
1046         Err(err) => {
1047             eprintln!("compute_next_event for '{}' failed - {}", event_str, err);
1048             return false;
1049         }
1050     };
1051
1052     let now = proxmox_time::epoch_i64();
1053     next <= now
1054 }
1055
1056 fn gather_disk_stats(disk_manager: Arc<DiskManage>, path: &Path, rrd_prefix: &str) {
1057
1058     match proxmox_backup::tools::disks::disk_usage(path) {
1059         Ok(status) => {
1060             let rrd_key = format!("{}/total", rrd_prefix);
1061             rrd_update_gauge(&rrd_key, status.total as f64);
1062             let rrd_key = format!("{}/used", rrd_prefix);
1063             rrd_update_gauge(&rrd_key, status.used as f64);
1064         }
1065         Err(err) => {
1066             eprintln!("read disk_usage on {:?} failed - {}", path, err);
1067         }
1068     }
1069
1070     match disk_manager.find_mounted_device(path) {
1071         Ok(None) => {},
1072         Ok(Some((fs_type, device, source))) => {
1073             let mut device_stat = None;
1074             match fs_type.as_str() {
1075                 "zfs" => {
1076                     if let Some(source) = source {
1077                         let pool = get_pool_from_dataset(&source).unwrap_or(&source);
1078                         match zfs_pool_stats(pool) {
1079                             Ok(stat) => device_stat = stat,
1080                             Err(err) => eprintln!("zfs_pool_stats({:?}) failed - {}", pool, err),
1081                         }
1082                     }
1083                 }
1084                 _ => {
1085                     if let Ok(disk) = disk_manager.clone().disk_by_dev_num(device.into_dev_t()) {
1086                         match disk.read_stat() {
1087                             Ok(stat) => device_stat = stat,
1088                             Err(err) => eprintln!("disk.read_stat {:?} failed - {}", path, err),
1089                         }
1090                     }
1091                 }
1092             }
1093             if let Some(stat) = device_stat {
1094                 let rrd_key = format!("{}/read_ios", rrd_prefix);
1095                 rrd_update_derive(&rrd_key, stat.read_ios as f64);
1096                 let rrd_key = format!("{}/read_bytes", rrd_prefix);
1097                 rrd_update_derive(&rrd_key, (stat.read_sectors*512) as f64);
1098
1099                 let rrd_key = format!("{}/write_ios", rrd_prefix);
1100                 rrd_update_derive(&rrd_key, stat.write_ios as f64);
1101                 let rrd_key = format!("{}/write_bytes", rrd_prefix);
1102                 rrd_update_derive(&rrd_key, (stat.write_sectors*512) as f64);
1103
1104                 let rrd_key = format!("{}/io_ticks", rrd_prefix);
1105                 rrd_update_derive(&rrd_key, (stat.io_ticks as f64)/1000.0);
1106             }
1107         }
1108         Err(err) => {
1109             eprintln!("find_mounted_device failed - {}", err);
1110         }
1111     }
1112 }
1113
1114 // Rate Limiter lookup
1115
1116 // Test WITH
1117 // proxmox-backup-client restore vm/201/2021-10-22T09:55:56Z drive-scsi0.img img1.img --repository localhost:store2
1118
1119 async fn run_traffic_control_updater() {
1120
1121      loop {
1122         let delay_target = Instant::now() +  Duration::from_secs(1);
1123
1124         {
1125             let mut cache = TRAFFIC_CONTROL_CACHE.lock().unwrap();
1126             cache.compute_current_rates();
1127         }
1128
1129          tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await;
1130      }
1131
1132 }
1133
1134 fn lookup_rate_limiter(
1135     peer: Option<std::net::SocketAddr>,
1136 ) -> (Option<Arc<dyn ShareableRateLimit>>, Option<Arc<dyn ShareableRateLimit>>) {
1137     let mut cache = TRAFFIC_CONTROL_CACHE.lock().unwrap();
1138
1139     let now = proxmox_time::epoch_i64();
1140
1141     cache.reload(now);
1142
1143     let (_rule_name, read_limiter, write_limiter) = cache.lookup_rate_limiter(peer, now);
1144
1145     (read_limiter, write_limiter)
1146 }